How soundwave-machine-learning/smlstarnet works
StarNet is a local-first desktop harness for creating AI-agent teams and running real work through them. The principal flow is: a person gives the station work; the frontend can route it through a spatial belt plan to an agent; the local sidecar runs the agent with model and tool access; completed files appear in the OUTBOX, while runtime state is projected in the station UI. The README documents these boundaries and capabilities, but the supplied source excerpts do not include the sidecar entry point or most frontend wiring, so uncertain call-level links are omitted.
Main entry points: sidecar Local sidecar · main.rs Desktop shell · workhub.js Work interface · app.js Station app
Components 23
Station UI
- Desktop shell
src-tauri/src/main.rs - Station app
frontend/app/app.js - Station world
frontend/app/world.js - Work interface
frontend/app/workhub.js - Deliverables
frontend/app/deliverables.js
Work orchestration
- Belt routing plan
frontend/app/pipeline.js - Agent setup
frontend/app/onboarding.js - Recipes and skills
frontend/app/recipes.js - Scheduled work
frontend/app/autojobs.js - Task briefs
frontend/app/intake.js
Agent runtime
- Local sidecar
sidecar - Model providers
sidecar - Agent tools
sidecar - MCP connectors
sidecar/mcp/manager.js - Message channels
frontend/app/channels.js - Voice interaction
frontend/app/voice.js
State and integrations
- Local persistence
frontend/app/save.js - Run and spend ledgers
frontend/app/recledger.js - Station projection
frontend/app/worldrenderer.js
Other components
- Station user
- Model services
- Chat services
- Local and network tools
Connections 26
- Station user → Desktop shell: opens
- Desktop shell → Local sidecar: starts
- Desktop shell → Station app: opens webview
- Station user → Station app: uses
- Station app → Work interface: presents work
- Work interface → Task briefs: captures task
- Work interface → Belt routing plan: configures routes
- Agent setup → Local sidecar: configures agents
- Recipes and skills → Local sidecar: launches work
- Scheduled work → Local sidecar: schedules runs
- Local sidecar → Belt routing plan: resolves target
- Local sidecar → Model providers: requests inference
- Model providers → Model services: calls models
- Local sidecar → Agent tools: dispatches tools
- Agent tools → Local and network tools: operates through
- Local sidecar → MCP connectors: uses connectors
- MCP connectors → Local and network tools: exposes tools
- Station user → Message channels: messages remotely
- Message channels → Local sidecar: delivers messages
- Station user → Voice interaction: dictates
- Local sidecar → Local persistence: persists state
- Local sidecar → Run and spend ledgers: records runs and spend
- Local sidecar → Deliverables: publishes deliverables
- Local sidecar → Station projection: emits runtime state
- Station projection → Station world: projects state
- Deliverables → Station user: offers files